基于反双曲正弦函数的抗冲激块稀疏自适应滤波算法

发布时间：2018-04-14 07:43

本文选题：自适应滤波器 + 非高斯噪声　；参考：《计算机应用》2017年01期

【摘要】：针对现有基于最小均方误差(MSE)的块稀疏系统辨识算法抗冲激性能不佳的问题,提出了一种利用反双曲正弦函数替代最小均方误差的改进型块稀疏归一化最小均方(IBS-NLMS)算法。该算法首先构造新的代价函数,利用负梯度最陡下降法求出增量,进而导出了新的滤波器权系数更新公式,在公式迭代过程中出现的冲激噪声会导致权系数的更新量趋于零向量,从而消除了由于非高斯冲激干扰而导致的算法发散问题。同时,理论分析并推导出了该算法的均值收敛过程。块稀疏系统辨识的仿真结果表明,在非高斯冲激噪声干扰和截断变化情况下,改进型算法与块稀疏归一化最小均方(BS-NLMS)算法相比有更快的收敛速度和更小的稳态误差。
[Abstract]:An improved block sparse normalized minimum mean square (IBS-NLMSS) algorithm using inverse hyperbolic sinusoidal function instead of minimum mean square error is proposed to solve the problem of poor impulse performance of existing block sparse system identification algorithms based on minimum mean square error (MSE).In this algorithm, a new cost function is constructed, and the increment is obtained by using the steepest descent method of negative gradient, and a new updating formula of filter weight coefficient is derived.The impulse noise in the iterative process of the formula will lead to the updating of the weight coefficient to zero vector, thus eliminating the problem of algorithm divergence caused by non- impulse interference.At the same time, the mean convergence process of the algorithm is analyzed and deduced theoretically.The simulation results of block sparse system identification show that the improved algorithm has faster convergence speed and smaller steady-state error than the block sparse normalized minimum mean square BS-NLMS algorithm in the case of non- impulse noise interference and truncation variation.
【作者单位】：信号与信息处理重庆市重点实验室(重庆邮电大学);
【基金】：国家自然科学基金资助项目(61501072) 重庆市科委自然科学基金资助项目(cstc2015jcyjA40027) 重庆邮电大学自然科学基金资助项目(A2015-60)~~
【分类号】：TN713

【相似文献】